Purchase this article with an account.
Irene Ruiz Hidalgo, Pablo Rodriguez Perez, Jos J Rozema, Carina Koppen, Sorcha S Ni Dhubhghaill, Nadia Zakaria, Marie-Jose B R Tassignon; Automated detection of Fuchs’ dystrophy through a machine learning algorithm using Pentacam data. Invest. Ophthalmol. Vis. Sci. 2015;56(7 ):1641.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
Fuchs’ dystrophy is a bilateral, age-related pathology of the corneal endothelium causing clinical symptoms, such as loss of visual acuity, guttae formation and increased central corneal thickness. Diagnosis is typically performed by slit lamp and specular microscopy evaluation, which is at times challenging in performance and interpretation. In this work we present the first results of a machine learning algorithm using only Pentacam topography/tomography data that may eventually assist physicians in the screening of Fuchs’ dystrophy.
<br /> This retrospective study includes data of 296 eyes, divided into two groups: 138 eyes diagnosed with Fuchs’ dystrophy (57.25% females and 42.75% males, mean age = 74.6 years, range 53 - 92 years), and 158 normal control eyes (63.29% females and 36.71% males, mean age = 63.3 years, range 50 - 91). Exclusion criteria were having a spherical equivalent refraction above 10 D, recent wear of hard contact lenses and previous corneal surgery. From the Pentacam we extracted 244 variables regarding densitometry, pachymetry and corneal volume, which were introduced into Weka, a computer program that supports data mining tasks. A support vector machine (SVM) algorithm was then run to obtain an automated binary classification of these eyes.
Precision, confusion matrix and area under the curve (AUC) were calculated for both groups and are shown in the table, along with the accuracy estimated through a 10-fold cross-validation (CV).
Pentacam data alone can provide a fairly good detection of Fuchs’ dystrophy and may therefore be useful alongside specular microscopy. Corneal densitometry parameters were found to be the most discriminant features, thus making these parameters good predictors of Fuchs’ dystrophy. In a later stage of the study subjects showing early signs of the disease could be included to assess the suitability of this automatic classification for early diagnosis.
This PDF is available to Subscribers Only